Add-ons that extend AI tools—such as "skills," plugins, GPTs, and extensions—can mimic well-known names to appear trustworthy, and once installed, they run with your permissions.
The AI tools used in companies today can be enhanced with small add-ons: “skills” for assistants, plugins, custom GPTs, and browser extensions. They’re convenient—they provide additional instructions to the tool and add new features. The question that few people ask is who wrote that add-on and what it actually does.
Security researchers have shown that some of these add-ons mimic well-known names—such as Anthropic and OpenAI—precisely to appear trustworthy and bypass the platform’s checks. A single wave of such add-ons containing hidden code garnered about 1.7 million installations. In one case, an add-on that downloaded videos falsely claimed to be affiliated with a well-known AI company.
The technical aspect is simple: once installed, an add-on runs with the privileges of the AI tool. It can read the data the AI has access to and execute commands on your behalf. No security breach or stolen password is needed—all it takes is for someone at the company to install the wrong add-on, lured by a name that seems trustworthy.
For a small company, the situation is clear: you weren’t hacked—you installed it yourself. And the decision about who can add extensions to AI tools usually falls between the cracks—no one takes responsibility, so anyone ends up doing it.
What you need to do:
- Install AI extensions and add-ons only from the manufacturer's official source, not from a link received via email or chat.
- Check the publisher before installing: a familiar name on the add-on doesn’t mean it belongs to them.
- Establish within the company who is authorized to add extensions, “skills,” or GPTs to AI work tools, and document the policy—even if it’s just a single paragraph.
- Remove add-ons that no one uses anymore—each one has the same permissions as the AI tool.
Sources:
This article was generated with AI assistance.